Find the check for the work in front of you: verify feed freshness, inspect agent actions, extract fields, route documents, evaluate search or compare an integration. Keep your own AI provider and systems.
The free feed-health, tool-policy, run-comparison, retrieval, classification, extraction, agent trace and webhook analyzers are separate and free; they are not included as MCP tools in Kit 1.1.0. Document Desk and the Kit are separate licenses; neither includes every tool on this site. Model, hosting and custom integration costs are not included in either $79 one-time price.
Preserve the baseline. Keep reference cases, judgments, model or rule versions, source revisions and contract snapshots in your own project.
Change one thing. Replay the same cases and retain failures, retries, abstentions and missing results.
Inspect the failure types. Look at individual cases, classes and cohorts before deciding whether a headline score improved the actual workflow.
Retain the evidence. Export the source-hashed report with run settings. Reports may include confidential field values, IDs and labels; review before sharing.